Image-Based Phenotyping Framework for Blackleg Disease in Canola: Progressing towards High-Throughput Analyses via Individual Plant Extraction

Abstract

Crop diseases are a significant constraint to agricultural production globally. Plant disease phenotyping is crucial for the identification, development, and deployment of effective breeding strategies, but phenotyping methodologies have not kept pace with the rapid progress in the genetic and genomic characterization of hosts and pathogens, still largely relying on visual assessment by trained experts. Remote sensing technologies were used to develop an automatic framework for extracting the stems of individual plants from RGB images for use in a pipeline for the automated quantification of blackleg crown canker (Leptopshaeria maculans) in mature Brassica napus plants. RGB images of the internal surfaces of stems cut transversely (cross-section) and vertically (longitudinal) were extracted from 722 and 313 images, respectively. We developed an image processing algorithm for extracting and spatially labeling up to eight individual plants within images. The method combined essential image processing techniques to achieve precise plant extraction. The approach was validated by performance metrics such as true and false positive rates and receiver operating curves. The framework was 98% and 86% accurate for cross-section and longitudinal sections, respectively. This algorithm is fundamental for the development of an accurate and precise quantification of disease in individual plants, with wide applications to plant research, including disease resistance and physiological traits for crop improvement.

Keywords

• binarization
• cropping
• morphological operations
• accuracy
• receiver operating curve